CN103471589A - Method for identifying walking mode and tracing track of pedestrian in room - Google Patents

Abstract

The invention discloses a method for identifying a walking mode and tracing a track of a pedestrian in a room. The method comprises the steps: firstly, defining a walking mode of the pedestrian in the room by using directions and speeds as the classifying basis according to the characteristic of a terrain under an indoor environment; then establishing a mathematical modeling for the indoor walking mode by applying a hidden Markov model (HMM) through using acquired acceleration sensor and geomagnetic sensor data; because of the interference of a metal object and the influence of body joggling in the walking process, a track primarily obtained by an indoor positioning system through dead reckoning possibly has large errors, and in the method, calculating the probability of each walking mode by using a viterbi algorithm of the HMM, and finding one group of walking modes with highest possibility; finally, matching an original location to a corresponding walking mode according to an identification result of the walking mode so as to correct an original track, and obtain a walking correcting track of the pedestrian in the room. According to the method, the precision of indoor positioning and track tracing can be remarkably improved.

Description

A kind of method of indoor pedestrian's walking mode identification and trajectory track

Technical field

The present invention relates to indoor pedestrian location and trajectory track field based on inertial sensor, relate in particular to that a kind of indoor pedestrian's walking mode based on implicit Markov model is identified and the method for trajectory track.

Background technology

Indoor positioning based on inertial sensor is under indoor environment, calculates step number and the course angle of pedestrian's walking by the data of the acceleration transducer that configures in the acquisition and processing mobile phone and geomagnetic sensor.In conjunction with the reckoning principle, on the basis of a position, according to the step number in the travel time section and course angle, extrapolate pedestrian's current location on known.Again by continuous seamless obtain pedestrian's current indoor location, can obtain pedestrian's track.

In pedestrian's trajectory track system based on inertial sensor, can calculate the step number of pedestrian's walking in conjunction with gait recognition method by gathering the acceleration transducer data.Pedestrian's course angle can be by the geomagnetic sensor data acquisition gathered.The step number that utilization is obtained and course angle can be obtained indoor pedestrian's position comparatively accurately in conjunction with the reckoning principle.But, because Geomagnetic signal is very easy to be subject to the impact in metal object and other geomagnetic noise sources, the geomagnetic sensor data of collection can produce larger error due to these chaff interferences.And the pedestrian while independently locating rocking also of health can produce and disturb the collection of geomagnetic sensor and acceleration transducer data.Moreover, these errors can be accumulated along with the growth of orientation distance, and this cumulative errors has affected indoor positioning based on inertial sensor and the precision of trajectory track system greatly.

In order to improve the precision of indoor pedestrian location and trajectory track system, the researchist proposed a lot of indoor orientation methods based on inertial sensor, what in them, have processes and filters inertial sensor data from macroscopic view, although reduced to a certain extent the impact of partly disturbing, fundamentally do not improved the precision of location.Some methods propose by adding for example WiFi of extra positioning equipment, and gyroscope etc. carry out the auxiliary positioning error of proofreading and correct.This method has significantly improved precision, but the extra positioning equipment added has increased substantially the cost of location, greatly reduces the availability of native system.

How under the prerequisite of not adding extras, successfully manage the interference of environmental factor and the impact of body-sway motion, accurately identify the abnormity point of indoor track, and revise the track of originally makeing mistakes, significantly improve the method for indoor positioning and trajectory track precision, be the advanced subject of industry research always.

Summary of the invention

Fundamental purpose of the present invention is to propose and realizes a kind of inertial sensor that utilizes mobile phone, carries out the identification of indoor pedestrian's walking mode, thereby realizes the method for indoor positioning and trajectory track accurately.The method can, under the prerequisite of not adding extras, successfully manage the interference of environmental factor and the impact of body-sway motion; The method can also accurately identify the abnormity point of indoor track, and the track abnormity point of originally makeing mistakes is revised, and significantly improves the precision of indoor positioning and trajectory track.

The technical solution adopted in the present invention is: a kind of method of indoor pedestrian's walking mode identification and trajectory track, it is characterized in that, and comprise the following steps:

Step 1: utilize acceleration transducer and geomagnetic sensor with adaptive sample frequency image data, obtain indoor pedestrian's initial position data;

Step 2: the more lower position data of utilizing this pedestrian in reckoning algorithm counting chamber;

Step 3: repeat described step 1 and step 2, by recording room one skilled in the art's position data constantly, obtain indoor this pedestrian's run trace;

Step 4: the implicit Markov model of setting up walking mode according to the characteristic of the general run trace of indoor common pedestrian;

Step 5: then the walking mode that identifies indoor this pedestrian according to the implicit Markov model of setting up and sensor observed reading matches indoor this pedestrian's run trace indoor this pedestrian's who identifies walking mode;

Step 6: adopted the method for abnormity point identification, deletion and bad block reparation to be revised the run trace after mating, obtained indoor this pedestrian's walking correction track.

As preferably, described in step 1, utilize acceleration transducer and geomagnetic sensor with adaptive sample frequency image data, its specific implementation comprises following sub-step:

Step 1.1: input three position angle h of pedestrian's walking constantly continuously _d, h _d-1, h _d-2; Wherein, d, d-1, d-2 is three continuous moment of carrying out data acquisition;

Step 1.2: calculate three continuous position angles between any two about the derivative of time

$h_{d-1}^{\′}=\frac{\mathrm{\Δh}}{\mathrm{\Δd}}=h_{d-1}-h_{d-2};$

Step 1.3: the second derivative between computer azimuth angle;

Step 1.4: judgement second derivative h _d' ' whether be less than minimum tolerance value ε;

If h _d' ' be less than ε, pedestrian's walking states is in stable state, and now degree of will speed up sensor and geomagnetic sensor are made as the low frequency image data;

If h _d' ' be greater than ε, pedestrian's walking states is in non-steady state, and now degree of will speed up sensor and geomagnetic sensor are made as the high-frequency image data.

As preferably, the more lower position data of utilizing reckoning algorithm counting chamber one skilled in the art described in step 2, its indoor pedestrian's more lower position is:

(x ^*,y ^*)=(x+sl×sc×cos?h,y+sl×sc+sin?h)；

Wherein: sl is pedestrian's step-length, and pedestrian's step-length obtains by estimation algorithm; Sc is the step number that the pedestrian walks, step number is that the Z axis acceleration figure obtained according to acceleration transducer calculates, in the process made a move due to pedestrian's normal row, it is a sine wave that the Z axis acceleration is worth curve, so the step number of by the number that detects Z axis acceleration sine wave, adding up the pedestrian; (x, y) is pedestrian's initial coordinate, (x ^*, y ^*) be the coordinate of pedestrian's more lower position; H is that the ，Gai position angle, position angle that the pedestrian walks obtains by geomagnetic sensor, and the position angle got is pedestrian's direction of advancing and the angle of magnetic north.

As preferably, the characteristic according to the general run trace of indoor common pedestrian described in step 4 is set up the implicit Markov model of walking mode, and its specific implementation comprises following sub-step:

Step 4.1: define indoor pedestrian's walking mode Hidden Markov Model (HMM), this model comprises these four parameters of implicit state, observed reading, state transition probability and output probability, and wherein, implicit state is corresponding to described indoor pedestrian's walking mode; Observed reading is the acceleration transducer of collection and the data of geomagnetic sensor; State transition probability is indoor occupant from a walking mode excessively to the probability of another one walking mode; Output probability is for a given observed reading, corresponding to the probability of certain walking states;

Step 4.2: according to described indoor pedestrian's walking characteristics, the Hidden Markov Model (HMM) of the indoor pedestrian's walking mode of instantiation.

As preferably, the Hidden Markov Model (HMM) of the indoor pedestrian's walking mode of the described instantiation of step 4.2, the instantiation method of the implicit state of the Hidden Markov Model (HMM) of its indoor pedestrian's walking mode is: at first indoor pedestrian's walking mode is classified according to direction of travel and the speed of travel; Wherein direction of travel is usingd to 15 degree as interval, 360 degree are divided into to 24 direction sections, the angle of supposing indoor pedestrian's direction of travel is θ, and the direction section under it is so

because of indoor normal walking speed scope, at 0.5m/s-2m/s, then the speed of travel being divided three classes, is respectively (0.5m/s-1m/s), middling speed (1m/s-1.5m/s), fast (1.5m/s-2m/s) at a slow speed; According to the direction of described walking and the speed of walking, walking mode can be divided into altogether 72 kinds, every kind of corresponding direction section of walking mode and a speed type.

As preferably, the Hidden Markov Model (HMM) of the indoor pedestrian's walking mode of the described instantiation of step 4.2, the instantiation method of the observed reading of the Hidden Markov Model (HMM) of its indoor pedestrian's walking mode is: by the pedestrian is divided into to several equal time intervals at the capable time made a move, gather an observed reading every an equal time period, described observed reading is arranged in to observation vector according to time sequencing; Wherein, described observed reading is three-dimensional vectorial O=(acce, Δ hd, spd), and acce means pedestrian's z-axle acceleration, the azimuthal variation amount that Δ hd means the pedestrian, the speed of travel that spd means the pedestrian.

As preferably, the Hidden Markov Model (HMM) of the indoor pedestrian's walking mode of the described instantiation of step 4.2, the instantiation method of the output probability of the Hidden Markov Model (HMM) of its indoor pedestrian's walking mode comprises following sub-step:

Step 4.2.1: the present invention is by capable corresponding three observational variables the method that it is averaged of making a move with each interior walking mode of multi collect indoor occupant, set up a standard fingerprint base for each walking mode, each fingerprint base comprises a walking mode s _jwith sequence of observations v _icorresponding relation, this corresponding relation is expressed as a bivector (s _j, v _i);

Step 4.2.2: a given sequence of observations v _i, search each walking mode s in fingerprint base _jcorresponding standard observation value sequence v _j;

Step 4.2.3: calculate given observed reading v _ithe standard observation value v corresponding with each walking mode _jbetween Euclidean distance;

Step 4.2.3: finally with the zero-mean Gaussian function, calculate each observed reading v _iwith corresponding state s _jbetween emission probability.

As preferably, the Hidden Markov Model (HMM) of the indoor pedestrian's walking mode of the described instantiation of step 4.2, the instantiation method of the state transition probability of the Hidden Markov Model (HMM) of its indoor pedestrian's walking mode is: described Hidden Markov Model (HMM) state transition probability is divided into direction transition probability and speed transition probability;

The direction transition probability means that the pedestrian transfers to the probability of the residing direction of another one state from a residing direction of state, if the pedestrian transfers to state j, i from state i _heading, j _headingmean respectively the residing direction section of state i and state j, direction transition probability P _headingthe computing formula of (i, j) is as follows:

$P_{\mathrm{heading}}(i,j)=\frac{\mathrm{\θ}}{2^n},n=|i_{\mathrm{heading}}-j_{\mathrm{heading}}|,$

Wherein θ means to work as i _heading=j _headingthe time the direction transition probability of pedestrian's Hidden Markov Model (HMM);

Because of P _heading(i, j) meets constraint condition

The probability sum of the direction that likely shifts be 1, therefore the value that by this constraint condition, can calculate θ is about 0.334; If the pedestrian transfers to state j, i from state i _speed, j _speedmean respectively state i and the residing speed type of state j, speed transition probability P _speedthe computing formula of (i, j) is as follows:

$P_{\mathrm{speed}}(i,j)=\frac{\mathrm{\ω}}{2^n},n=|i_{\mathrm{speed}}-j_{\mathrm{speed}}|,$

Wherein ω means to work as i _speed=j _speedthe time the direction transition probability of pedestrian's Hidden Markov Model (HMM);

Because of P _speed(i, j) meets following constraint condition

$\underset{j=1}{\overset{k}{\mathrm{\Σ}}}{P}_{\mathrm{speed}}(i,j)=1,$

The probability sum of the speed that likely shifts be 1, therefore can calculate ω by this constraint condition, be 0.5;

, for a Hidden Markov Model (HMM) λ=(A, B, π), state transition probability matrix A is:

a _ij＝P _heading(i，j)×P _speed(i，j)，

Wherein, a _ijmean the state transition probability of Hidden Markov Model (HMM), B means the output probability matrix of Hidden Markov Model (HMM), and π means the original state probability matrix of Hidden Markov Model (HMM).

As preferably, the walking mode that identifies indoor this pedestrian according to the implicit Markov model of setting up and sensor observed reading described in step 5, its specific implementation comprises following sub-step:

Step 5.1: the sequence of observations O that inputs any walking mode _i, i=1,2 ... n;

Step 5.2: set up Hidden Markov Model (HMM) λ=(A, B, the π) of indoor pedestrian's walking mode, wherein:

π＝[π ₁，π ₂，...，π _N]；

Wherein, a _ijthe state transition probability that means Hidden Markov Model (HMM), b _j(i) mean the output probability of Hidden Markov Model (HMM), N means the status number of implicit Er Kefu model, π _jthe original state probability that means Hidden Markov Model (HMM);

Step 5.3: Hidden Markov Model (HMM) λ=(A, B, the π) for pedestrian's walking mode of setting up, calculate with viterbi algorithm the probability that all possible walking states is corresponding, then by sequence, find out maximum probability wherein:

δ ₁(j)=π _jb _j(o ₁) 1≤j≤N

δ _t(j)=max _i(δ _t-1(j)a _ijb _i(o _t)) 2≤t≤T,1≤k≤N；

Wherein, T means the time of the walking cost of this walking mode, calculates t constantly in state S _jmaximum probability δ t (j);

Step 5.4: judgement t is the state S in maximum probability constantly _jprobability δ t (j) be greater than fixing threshold value p (TH)?

If: continue to carry out following step 5.5,

If not: finish identification;

Step 5.5: state corresponding to maximum probability that record calculates at every turn, for recalling state each time; Maximum probability δ for a viterbi algorithm computation process _t-1(i) and one state transition probability a _ij, traceback information ψ _t(j) computing formula is as follows:

ψ _t(j)=arg?max _i[δ _t-1(i)a _ij] 2≤t≤T,1≤j≤N；

Step 5.6: utilize described traceback information ψ _t(j), with back-track algorithm, identify pedestrian's walking mode $q_t^{*},t=T-1,T-2,...,1,$ Its iterative process is as follows:

$q_T^{*}=S_N$

$q_t^{*}={\mathrm{\ψ}}_{t+1}\left(q_{t+1}^{*}\right),t=T-1,T-2,...,1.$

As preferably, described p (TH) gets 0.6.

As preferably, the employing described in step 6 method of abnormity point identification, deletion and the bad block reparation run trace after to coupling revised, its specific implementation comprises following sub-step:

Step 6.1: the sequence of observations O that inputs any one walking mode _i, i=1,2 ... n; Hidden Markov Model (HMM) λ=(A, B, the π) of indoor pedestrian's walking mode of setting up; Wherein, A means the state transition probability matrix of Hidden Markov Model (HMM), and B means the output probability matrix of Hidden Markov Model (HMM), and π means the original state probability matrix of Hidden Markov Model (HMM).

Step 6.2: set a moving window that length is k for the sequence of observations, the sequence of observations that is n by length is divided into the short sequence that length is k;

Step 6.3: moving window moves a unit at every turn backward, wherein, and the number of times C=(n-k+1) that moving window moves altogether;

Step 6.4: judge whether moving window arrives sequence of observations end,

If so, monitoring finishes;

If not, continue to carry out following step;

Step 6.5: the maximum probability δ t (j) of walking mode that calculates the short sequence at this moving window place with viterbi algorithm:

δ ₁(j)=π _jb _j(O ₁) 1≤j≤N

${\mathrm{\δ}}_t\left(j\right)={\max }_i\left({\mathrm{\δ}}_{t-1}\left(j\right)a_{\mathrm{ij}}{b}_{{\mathrm{iO}}_t}\right),2\≤t\≤T,1\≤k\≤N;$

Wherein, a _ijthe state transition probability that means Hidden Markov Model (HMM), b _j(i) mean the output probability of Hidden Markov Model (HMM), N means the status number of implicit Er Kefu model, π _jthe original state probability that means Hidden Markov Model (HMM), T means the time of the walking cost of this walking mode;

Step 6.6: judgement maximum probability δ t (j) and threshold values p (TH) size of setting;

If maximum rating probability δ t (j) is greater than this threshold value, this time short sequence is normal point, returns and continues to carry out described step 6.3;

If maximum rating probability δ t (j) is less than this threshold value, this time short sequence is abnormity point, continues to carry out following step;

Step 6.7: this short sequence is carried out to abnormity point and delete and the bad block reparation, the method for its bad block reparation is first to delete original abnormity point, then passes through the method for curve point of position matching in the abnormity point of deleting, then returns and carry out described step 6.3.

The present invention has following innovation point:

(1) in traditional indoor positioning technology, generally adopt the location fingerprint location, or adopt based on signal attenuation distance model localization method, these methods are not only located large but also positioning precision is lower; In recent years, the researchist proposed to adopt the indoor orientation method based on inertial sensor, and this is a kind of autonomous, continuous indoor positioning technology, and positioning precision is high and stable.But the indoor locating system that is based on inertial sensor works long hours and can produce cumulative errors, greatly the precision of impact location; For the cumulative errors that the location technology reduced based on inertial sensor exists, the present invention, from the angle of microcosmic, identifies the slight change that indoor pedestrian often makes a move;

(2) in the indoor positioning technology based on inertial sensor, sensing data easily is subject to the interference of environment, and self also has fluctuation, and these factors can cause the type one skilled in the art's of calculating the serious deviation of trajectory generation.Technology before is to adopt the method for filtered sensor data to eliminate exceptional value, although this method has been eliminated some errors to a certain extent, does not fundamentally solve the problem of geomagnetic noise and shake.The technology also had is carried out correction error by adding extra auxiliary positioning equipment, although this method has improved positioning precision, need to pay higher location cost.The present invention, by set up the fine-grained model based on implicit Markov for the walking mode of indoor occupant, identifies the pattern of indoor walking, thereby revises the track of walking.Experimental results show that the method has significantly improved the precision of indoor positioning;

(3) this transmission adopts adaptive sensor data acquisition frequency, can dynamically adjust the frequency of sampling according to different indoor environments, than the collecting method of fixed frequency, has significantly lowered the energy consumption of sensor;

(4) way that the present invention proposes is simple and easy to realize, without any need for extra positioning equipment, and very high robustness and reliability are arranged, be applicable to various indoor environments.

The accompanying drawing explanation

Fig. 1: be method flow diagram of the present invention.

Fig. 2: be the method flow diagram of adaptive frequency data acquisition of the present invention.

Fig. 3: be reckoning principle schematic of the present invention.

Fig. 4: be the implicit view of the implicit Markov model of walking mode of the present invention.

Fig. 5: be the method flow diagram of walking mode identification of the present invention.

Fig. 6: be the method flow diagram of abnormity point identification of the present invention, deletion and bad block reparation.

Embodiment

Below with reference to the drawings and specific embodiments, the present invention is further elaborated.

Ask for an interview Fig. 1, the technical solution adopted in the present invention is: a kind of method of indoor pedestrian's walking mode identification and trajectory track comprises the following steps:

Step 1: utilize acceleration transducer and geomagnetic sensor with adaptive sample frequency image data, obtain indoor pedestrian's initial position data;

Step 2: the more lower position data of utilizing this pedestrian in reckoning algorithm counting chamber;

Step 3: repeated execution of steps 1 and step 2, by recording room one skilled in the art's position data constantly, obtain indoor this pedestrian's run trace;

Step 4: the implicit Markov model of setting up walking mode according to the characteristic of the general run trace of indoor common pedestrian;

Step 5: then the walking mode that identifies indoor this pedestrian according to the implicit Markov model of setting up and sensor observed reading matches indoor this pedestrian's run trace indoor this pedestrian's who identifies walking mode;

Step 6: adopted the method for abnormity point identification, deletion and bad block reparation to be revised the run trace after mating, obtained indoor this pedestrian's walking correction track.

Ask for an interview Fig. 2, in step 1, utilize acceleration transducer and geomagnetic sensor with adaptive sample frequency image data, its specific implementation comprises following sub-step:

Step 1.1: input three position angle h of pedestrian's walking constantly continuously _d, h _d-1, h _d-2; Wherein, d, d-1, d-2 is three continuous moment of carrying out data acquisition;

Step 1.2: calculate three continuous position angles between any two about the derivative of time

$h_{d-1}^{\′}=\frac{\mathrm{\Δh}}{\mathrm{\Δd}}=h_{d-1}-h_{d-2};$

Step 1.3: the second derivative between computer azimuth angle;

Step 1.4: judgement second derivative h _d' ' whether be less than minimum tolerance value ε;

If h _d' ' be less than ε, pedestrian's walking states is in stable state, and now degree of will speed up sensor and geomagnetic sensor are made as the low frequency image data;

If h _d' ' be greater than ε, pedestrian's walking states is in non-steady state, and now degree of will speed up sensor and geomagnetic sensor are made as the high-frequency image data.

Ask for an interview Fig. 3, utilize reckoning algorithm counting chamber one skilled in the art's more lower position data in step 2, its indoor pedestrian's more lower position is:

(x ^*,y ^*)=(x+sl×sc×cos?h,y+sl×sc+sin?h)；

Wherein: sl is pedestrian's step-length, and pedestrian's step-length obtains by estimation algorithm; Sc is the step number that the pedestrian walks, step number is that the Z axis acceleration figure obtained according to acceleration transducer calculates, in the process made a move due to pedestrian's normal row, it is a sine wave that the Z axis acceleration is worth curve, so the step number of by the number that detects Z axis acceleration sine wave, adding up the pedestrian; (x, y) is pedestrian's initial coordinate, (x ^*, y ^*) be the coordinate of pedestrian's more lower position; H is that the ，Gai position angle, position angle that the pedestrian walks obtains by geomagnetic sensor, and the position angle got is pedestrian's direction of advancing and the angle of magnetic north.

Set up the implicit Markov model of walking mode in step 4 according to the characteristic of the general run trace of indoor common pedestrian, its specific implementation comprises following sub-step:

Step 4.1: define indoor pedestrian's walking mode Hidden Markov Model (HMM), this model comprises these four parameters of implicit state, observed reading, state transition probability and output probability, and wherein, implicit state is corresponding to indoor pedestrian's walking mode; Observed reading is the acceleration transducer of collection and the data of geomagnetic sensor; State transition probability is indoor occupant from a walking mode excessively to the probability of another one walking mode; Output probability is for a given observed reading, corresponding to the probability of certain walking states;

Step 4.2: according to indoor pedestrian's walking characteristics, the Hidden Markov Model (HMM) of the indoor pedestrian's walking mode of instantiation; The instantiation method of the implicit state of the Hidden Markov Model (HMM) of its indoor pedestrian's walking mode is: at first indoor pedestrian's walking mode is classified according to direction of travel and the speed of travel; Wherein direction of travel is usingd to 15 degree as interval, 360 degree are divided into to 24 direction sections, the angle of supposing indoor pedestrian's direction of travel is θ, and the direction section under it is so

because of indoor normal walking speed scope, at 0.5m/s-2m/s, then the speed of travel being divided three classes, is respectively (0.5m/s-1m/s), middling speed (1m/s-1.5m/s), fast (1.5m/s-2m/s) at a slow speed; According to the direction of walking and the speed of walking, walking mode can be divided into altogether 72 kinds, every kind of corresponding direction section of walking mode and a speed type;

The instantiation method of the observed reading of the Hidden Markov Model (HMM) of its indoor pedestrian's walking mode is: by the pedestrian is divided into to several equal time intervals at the capable time made a move, gather an observed reading every an equal time period, observed reading is arranged in to observation vector according to time sequencing; Wherein, observed reading be three-dimensional vectorial O=(acce, Δ hd, spd), acce means pedestrian's z-axle acceleration, the azimuthal variation amount that Δ hd means the pedestrian, the speed of travel that spd means the pedestrian;

The instantiation method of the output probability of the Hidden Markov Model (HMM) of its indoor pedestrian's walking mode comprises following sub-step:

Step 4.2.1: the present invention is by capable corresponding three observational variables the method that it is averaged of making a move with each interior walking mode of multi collect indoor occupant, set up a standard fingerprint base for each walking mode, each fingerprint base comprises a walking mode s _jwith sequence of observations v _icorresponding relation, this corresponding relation is expressed as a bivector (s _j, v _i);

Step 4.2.2: the sequence of observations v of given HMM _ican be expressed as v _i=[O ₁, O ₂..., O _n], find corresponding states s by searching fingerprint base _jthe standard value sequence

calculating observation value sequence v _iand v _jbetween Euclidean distance, known according to the definition of observed reading, for sequence of observations v _i, v _jin each observed quantity O _nand O _n ^*, n=1,2 ... N,, be all a tri-vector that comprises three observational componentses;

O _n=(acce _n,Δhd _n,spd _n)，

O _n ^*=(acce _n ^*,Δhd _n ^*,spd _n ^*),n=1,2,…N

O _nand O _n ^*between Euclidean distance dis (O _n, O _n ^*) computing method are as follows:

$\mathrm{dis}(O_n,O_n^{*})=\sqrt{{({\mathrm{acce}}_n-{{\mathrm{acce}}_n}^{*})}^2+{({\mathrm{\Δhd}}_n-{{\mathrm{\Δhd}}_n}^{*})}^2+{({\mathrm{apd}}_n-{{\mathrm{apd}}_n}^{*})}^2}$

So given observed reading v _iwith standard observation value v _jbetween Euclidean distance can be expressed as:

$\begin{array}{c}\mathrm{dis}(v_i,v_j)=\sqrt{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}\mathrm{dis}{(O_n,O_n^{*})}^2}\\ =\sqrt{\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{({\mathrm{acce}}_n-{{\mathrm{acce}}_n}^{*})}^2+{({\mathrm{\Δhd}}_n-{{\mathrm{\Δhd}}_n}^{*})}^2+{({\mathrm{spd}}_n-{{\mathrm{spd}}_n}^{*})}^2}\end{array}$

Step 4.2.3: to Euclidean distance dis (v _i, v _j) ask the zero-mean Gaussian function to calculate each observed reading v _iwith corresponding state s _jbetween emission probability, formula is as follows:

$b_j\left(i\right)=P(o_t=v_i|q_t=s_j)=g\left(\mathrm{dis}(v_i,v_j)\right)=e^{\frac{{-\mathrm{dis}(v_i-v_j)}^2}{{2\mathrm{\δ}}^2}};$

The instantiation method of the state transition probability of the Hidden Markov Model (HMM) of its indoor pedestrian's walking mode is: the Hidden Markov Model (HMM) state transition probability is divided into direction transition probability and speed transition probability;

The direction transition probability means that the pedestrian transfers to the probability of the residing direction of another one state from a residing direction of state, if the pedestrian transfers to state j, i from state i _heading, j _headingmean respectively the residing direction section of state i and state j, direction transition probability P _headingthe computing formula of (i, j) is as follows:

$P_{\mathrm{heading}}(i,j)=\frac{\mathrm{\θ}}{2^n},n=|i_{\mathrm{heading}}-j_{\mathrm{heading}}|,$

Wherein θ means to work as i _heading=j _headingthe time the direction transition probability of pedestrian's Hidden Markov Model (HMM);

Because of P _heading(i, j) meets constraint condition

The probability sum of the direction that likely shifts be 1, therefore the value that by this constraint condition, can calculate θ is about 0.334; If the pedestrian transfers to state j, i from state i _speed, j _speedmean respectively state i and the residing speed type of state j, speed transition probability P _speedthe computing formula of (i, j) is as follows:

$P_{\mathrm{speed}}(i,j)=\frac{\mathrm{\ω}}{2^n},n=|i_{\mathrm{speed}}-j_{\mathrm{speed}}|,$

Wherein ω means to work as i _speed=j _speedthe time the direction transition probability of pedestrian's Hidden Markov Model (HMM);

Because of P _speed(i, j) meets following constraint condition

$\underset{j=1}{\overset{k}{\mathrm{\Σ}}}{P}_{\mathrm{speed}}(i,j)=1,$

The probability sum of the speed that likely shifts be 1, therefore can calculate ω by this constraint condition, be 0.5;

, for a Hidden Markov Model (HMM) λ=(A, B, π), state transition probability matrix A is:

a _ij＝P _heading(i，j)×P _speed(i，j)，

Wherein, a _ijmean the state transition probability of Hidden Markov Model (HMM), B means the output probability matrix of Hidden Markov Model (HMM), and π means the original state probability matrix of Hidden Markov Model (HMM).

Ask for an interview Fig. 4, the implicit Markov model HMM schematic diagram of instantiation walking mode, it is a kind ofly with probabilistic method, to describe non-static state (be the statistics of signal over time) signal source model.HMM is comprised of two correlated processes:

(1) inherence, sightless Markov chain, comprise limited state, state-transition matrix and an original state probability distribution;

(2) the output probability density function collection be associated with each state of Markov chain.

Usually, a HMM can mean by following tlv triple:

λ=(A,B,π)；

The hidden Markov process characteristic can be by the incompatible sign of following parameter set:

(1) implicit state (hidden states), the implicit state key reaction in implicit Markov model the residing state of object, in the present invention, be exactly indoor pedestrian's walking mode.Ask for an interview Fig. 4, defined the pattern of indoor walking according to characteristics the present invention of indoor pedestrian's walking.The present invention classifies indoor walking mode according to direction of travel and the speed of travel.Wherein direction of travel is usingd to 15 degree as interval, 360 degree are divided into to 24 direction sections, if the angle of direction of travel is θ, the direction section under it is so

due to indoor normal walking speed scope, greatly about 0.5m/s-2m/s, the present invention can be divided three classes the speed of travel accordingly, is respectively (0.5m/s-1m/s), middling speed (1m/s-1.5m/s), fast (1.5m/s-2m/s) at a slow speed.According to direction and the speed of walking, walking mode can be divided into altogether 72 kinds, direction section of every kind of correspondence and a speed type so;

(2) observed reading (observables) is in order to reflect the characteristics of indoor pedestrian's walking comprehensively, the present invention chooses the z axle acceleration of pedestrian's acceleration transducer in the capable time made a move, and the azimuthal variation amount of geomagnetic sensor and the speed of travel are as observed reading.Observed reading can be expressed as the vectorial O=(acce, Δ hd, spd) of a three-dimensional, means respectively z-axle acceleration that acceleration sensing records, pedestrian's that geomagnetic sensor records azimuthal variation amount and pedestrian's the speed of travel.Select the z axle acceleration to be because indoor pedestrian when the speed with different and direction walking, the amplitude that health teetertotters is also different, and the z axle acceleration can reflect the variation of the acceleration of pedestrian's vertical direction fully.Two amounts in the corresponding implicit state of position angle and speed, position angle can record by geomagnetic sensor, and the speed of travel can calculate by pedestrian's estimating step length and walking cycle.The time that the present invention makes a move row is divided into several equal time intervals, every an equal time period, gathers an observed reading, then observed reading is arranged in to observation vector according to time sequencing.For example, suppose that the time that pedestrian's row makes a move is T, the time that row is made a move is divided into N the equal time interval, and the moment of pick-up transducers data is so

the sensing data constantly collected at i is O _i.The observation sequence v gathered so _kcan be expressed as one by O _ithe N dimension group v formed _k=[O ₁, O ₂... O _n].For each observed reading O _i, the observed reading of the capable time that makes a move of indoor pedestrian of choosing comprises: pedestrian's z-axle acceleration, pedestrian's azimuthal variation amount and pedestrian's the speed of travel, observed reading O so _ican be expressed as a tri-vector O _i=(acce, Δ hd, spd);

(3) original state probability: refer to first state q ₁get s=[s actually ₁, s ₂..., s _n] in which probability, it forms 1 * N vector π, π _i=p (q ₁=s _i) and π=[π ₁, π ₂..., π _n], we set π ₁=1 (π _i=0, i ≠ 1), suppose that HMM is from first state;

(4) transition probability (transition probability) a _ij=P (q _t+1=s _j| q _t=s _i): by state s _itransfer to state s _jprobability (to first-order Markov process).Due to the total possible state of N kind, therefore total N * N possible value.They are become to A=[a with matrix formulation _ij] and

the present invention is divided into direction transition probability p by state transition probability _headingwith speed transition probability p _speed;

The direction transition probability means that the pedestrian transfers to the probability of the residing direction of another one state from a residing direction of state, if the pedestrian transfers to state j, i from state i _heading, j _headingmean respectively the residing direction section of state i and state j, direction transition probability P _headingthe computing formula of (i, j) is as follows:

$P_{\mathrm{heading}}(i,j)=\frac{\mathrm{\θ}}{2^n},n=|i_{\mathrm{heading}}-j_{\mathrm{heading}}|,$

Wherein θ means to work as i _heading=j _headingthe time the direction transition probability of pedestrian's Hidden Markov Model (HMM);

Because of P _heading(i, j) meets constraint condition

The probability sum of the direction that likely shifts be 1, therefore the value that by this constraint condition, can calculate θ is about 0.334;

If the pedestrian transfers to state j, i from state i _speed, j _speedmean respectively state i and the residing speed type of state j, speed transition probability P _speedthe computing formula of (i, j) is as follows:

$P_{\mathrm{speed}}(i,j)=\frac{\mathrm{\ω}}{2^n},n=|i_{\mathrm{speed}}-j_{\mathrm{speed}}|,$

Wherein ω means to work as i _speed=j _speedthe time the direction transition probability of pedestrian's Hidden Markov Model (HMM);

Because of P _speed(i, j) meets following constraint condition

$\underset{j=1}{\overset{k}{\mathrm{\Σ}}}{P}_{\mathrm{speed}}(i,j)=1$

The probability sum of the speed that likely shifts be 1, therefore can calculate ω by this constraint condition, be 0.5;

, for a Hidden Markov Model (HMM) λ=(A, B, π), state transition probability matrix A is:

a _ij＝P _heading(i，j)×P _speed(i，j)，

Wherein, a _ijmean the state transition probability of Hidden Markov Model (HMM), B means the output probability matrix of Hidden Markov Model (HMM), and π means the original state probability matrix of Hidden Markov Model (HMM);

(5) output probability (emission probability) b _j(i)=P (o _t=v _i| q _t=s _j): at state s _jv is observed in lower generation _iprobability.If possible observation in total M, form N * Metzler matrix B=[b _ij] and

in order to weigh the degree of closeness of each sequence of observations and each state, the present invention is by capable corresponding three observational variables the method that it is averaged of making a move with each interior walking mode of multi collect indoor occupant, set up a standard fingerprint base for each walking mode, each fingerprint base comprises a walking states s _jwith sequence of observations v _icorresponding relation, fingerprint base can be expressed as a bivector (s _j, v _i).The sequence of observations vi of HMM can be expressed as v _i=[O ₁, O ₂..., O _n], corresponding states s in fingerprint base _jthe standard value sequence be

according to above the definition about observed reading is known, for each the observed quantity O in the sequence of observations _nand O _n ^*, n=1,2 ... N,, be all a tri-vector that comprises three observational componentses.

On=(acce _n,Δhd _n,spd _n)，

On ^*=(acce _n ^*,Δhd _n ^*,spd _n ^*),n=1,2,…N

The present invention adopts Euclidean distance to weigh two distances between variable, On and On ^*between Euclidean distance dis (O _n, O _n ^*) computing method are as follows:

Euclidean distance between so given observed reading vi and standard observation value vj can be expressed as:

Finally use Euclidean distance dis (v _i, v _j) ask the method for zero-mean Gaussian function to calculate the output probability between each observed reading vi and corresponding state sj, formula is as follows:

$b_j\left(i\right)=P(o_t=v_i|q_t=s_j)=g\left(\mathrm{dis}(v_i,v_j)\right)=e^{\frac{{-\mathrm{dis}(v_i-v_j)}^2}{{2\mathrm{\δ}}^2}}.$

Ask for an interview Fig. 5, the walking mode that identifies indoor this pedestrian according to the implicit Markov model of setting up and sensor observed reading in step 5, its specific implementation comprises following sub-step:

Step 5.1: the sequence of observations O that inputs any walking mode _i, i=1,2 ... n;

Step 5.2: set up Hidden Markov Model (HMM) λ=(A, B, the π) of indoor pedestrian's walking mode wherein

π＝[π ₁，π ₂，...，π _N]；

Wherein, a _ijthe state transition probability that means Hidden Markov Model (HMM), b _j(i) mean the output probability of Hidden Markov Model (HMM), N means the status number of implicit Er Kefu model, π _jthe original state probability that means Hidden Markov Model (HMM);

Step 5.3: Hidden Markov Model (HMM) λ=(A, B, the π) for pedestrian's walking mode of setting up, calculate with viterbi algorithm the probability that all possible walking states is corresponding, then by sequence, find out maximum probability wherein:

δ ₁(j)=π _jb _j(o ₁) 1≤j≤N

δ _t(j)=max _i(δ _t-1(j)a _ijb _i(o _t)) 2≤t≤T,1≤k≤N；

Wherein, T means the time of the walking cost of this walking mode, calculates t constantly in state S _jmaximum probability δ t (j);

Step 5.4: judgement t is the state S in maximum probability constantly _jprobability δ t (j) be greater than fixing threshold value p (TH)?

If: continue to carry out following step 5.5,

If not: finish identification;

Wherein, the p of the present embodiment (TH) gets 0.6;

Step 5.5: state corresponding to maximum probability that record calculates at every turn, for recalling state each time; Maximum probability δ for a viterbi algorithm computation process _t-1(i) and one state transition probability a _ij, traceback information ψ _t(j) computing formula is as follows:

ψ _t(j)=arg?max _i[δ _t-1(i)a _ij] 2≤t≤T,1≤j≤N；

Step 5.6: utilize traceback information ψ _t(j), with back-track algorithm, identify pedestrian's walking mode its iterative process is as follows:

$q_T^{*}=S_N$

$q_t^{*}={\mathrm{\ψ}}_{t+1}\left(q_{t+1}^{*}\right),t=T-1,T-2,...,1.$

Ask for an interview Fig. 6, in step 6, adopted the method for abnormity point identification, deletion and bad block reparation to be revised the run trace after mating, its specific implementation comprises following sub-step:

Step 6.1: the sequence of observations O that inputs any one walking mode _i, i=1,2 ... n; Hidden Markov Model (HMM) λ=(A, B, the π) of indoor pedestrian's walking mode of setting up; Wherein, A means the state transition probability matrix of Hidden Markov Model (HMM), and B means the output probability matrix of Hidden Markov Model (HMM), and π means the original state probability matrix of Hidden Markov Model (HMM).

Step 6.2: set a moving window that length is k for the sequence of observations, the sequence of observations that is n by length is divided into the short sequence that length is k;

Step 6.3: moving window moves a unit at every turn backward, wherein, and the number of times C=(n-k+1) that moving window moves altogether;

Step 6.4: judge whether moving window arrives sequence of observations end,

If so, monitoring finishes;

If not, continue to carry out following step;

Step 6.5: the maximum probability δ t (j) of walking mode that calculates the short sequence at this moving window place with viterbi algorithm:

δ ₁(j)=π _jb _j(O ₁) 1≤j≤N

${\mathrm{\δ}}_t\left(j\right)={\max }_i\left({\mathrm{\δ}}_{t-1}\left(j\right)a_{\mathrm{ij}}{b}_{{\mathrm{iO}}_t}\right),2\≤t\≤T,1\≤k\≤N;$

Wherein, a _ijthe state transition probability that means Hidden Markov Model (HMM), b _j(i) mean the output probability of Hidden Markov Model (HMM), N means the status number of implicit Er Kefu model, π _jthe original state probability that means Hidden Markov Model (HMM), T means the time of the walking cost of this walking mode;

Step 6.6: judgement maximum probability δ t (j) and threshold values p (TH) size of setting;

If maximum rating probability δ t (j) is greater than this threshold value, this time short sequence is normal point, returns and continues execution step 6.3;

If maximum rating probability δ t (j) is less than this threshold value, this time short sequence is abnormity point, continues to carry out following step;

Step 6.7: this short sequence is carried out to abnormity point and delete and the bad block reparation, the method for its bad block reparation is first to delete original abnormity point, then passes through the method for curve point of position matching in the abnormity point of deleting, then returns to execution step 6.3.

Above specific embodiment has been described in detail purpose of the present invention, technical scheme and beneficial effect.Institute's foregoing that it should be understood that, only for specific embodiments of the invention, is not limited to the present invention.All within spirit of the present invention and principle, any modification of making, be equal to and replace and improvement etc., within all should being included in protection scope of the present invention.

Claims (11)

1. the method for indoor pedestrian's walking mode identification and trajectory track, is characterized in that, comprises the following steps:

Step 1: utilize acceleration transducer and geomagnetic sensor with adaptive sample frequency image data, obtain indoor pedestrian's initial position data;

Step 2: the more lower position data of utilizing this pedestrian in reckoning algorithm counting chamber;

Step 3: repeat described step 1 and step 2, by recording room one skilled in the art's position data constantly, obtain indoor this pedestrian's run trace;

Step 4: the implicit Markov model of setting up walking mode according to the characteristic of the general run trace of indoor common pedestrian;

Step 5: then the walking mode that identifies indoor this pedestrian according to the implicit Markov model of setting up and sensor observed reading matches indoor this pedestrian's run trace indoor this pedestrian's who identifies walking mode;

Step 6: adopted the method for abnormity point identification, deletion and bad block reparation to be revised the run trace after mating, obtained indoor this pedestrian's walking correction track.

2. the method for indoor pedestrian's walking mode identification according to claim 1 and trajectory track, it is characterized in that: described in step 1, utilize acceleration transducer and geomagnetic sensor with adaptive sample frequency image data, its specific implementation comprises following sub-step:

Step 1.1: input three position angle h of pedestrian's walking constantly continuously _d, h _d-1, h _d-2; Wherein, d, d-1, d-2 is three continuous moment of carrying out data acquisition;

Step 1.2: calculate three continuous position angles between any two about the derivative of time

$h_{d-1}^{\′}=\frac{\mathrm{\Δh}}{\mathrm{\Δd}}=h_{d-1}-h_{d-2};$

Step 1.3: the second derivative between computer azimuth angle;

Step 1.4: judgement second derivative h _d' ' whether be less than minimum tolerance value ε;

If h _d' ' be less than ε, pedestrian's walking states is in stable state, and now degree of will speed up sensor and geomagnetic sensor are made as the low frequency image data;

If h _d' ' be greater than ε, pedestrian's walking states is in non-steady state, and now degree of will speed up sensor and geomagnetic sensor are made as the high-frequency image data.

3. the method for indoor pedestrian's walking mode identification according to claim 1 and trajectory track, it is characterized in that: the more lower position data of utilizing reckoning algorithm counting chamber one skilled in the art described in step 2, its indoor pedestrian's more lower position is: (x ^*, y ^*)=(x+sl * sc * cos h, y+sl * sc+sin h);

Wherein: sl is pedestrian's step-length, and pedestrian's step-length obtains by estimation algorithm; Sc is the step number that the pedestrian walks, step number is that the Z axis acceleration figure obtained according to acceleration transducer calculates, in the process made a move due to pedestrian's normal row, it is a sine wave that the Z axis acceleration is worth curve, so the step number of by the number that detects Z axis acceleration sine wave, adding up the pedestrian; (x, y) is pedestrian's initial coordinate, (x ^*, y ^*) be the coordinate of pedestrian's more lower position; H is that the ，Gai position angle, position angle that the pedestrian walks obtains by geomagnetic sensor, and the position angle got is pedestrian's direction of advancing and the angle of magnetic north.

4. the method for indoor pedestrian's walking mode identification according to claim 1 and trajectory track, it is characterized in that: the characteristic according to the general run trace of indoor common pedestrian described in step 4 is set up the implicit Markov model of walking mode, and its specific implementation comprises following sub-step:

Step 4.1: define indoor pedestrian's walking mode Hidden Markov Model (HMM), this model comprises these four parameters of implicit state, observed reading, state transition probability and output probability, and wherein, implicit state is corresponding to described indoor pedestrian's walking mode; Observed reading is the acceleration transducer of collection and the data of geomagnetic sensor; State transition probability is indoor occupant from a walking mode excessively to the probability of another one walking mode; Output probability is for a given observed reading, corresponding to the probability of certain walking states;

Step 4.2: according to described indoor pedestrian's walking characteristics, the Hidden Markov Model (HMM) of the indoor pedestrian's walking mode of instantiation.

5. the method for indoor pedestrian's walking mode identification according to claim 4 and trajectory track, it is characterized in that: the Hidden Markov Model (HMM) of the indoor pedestrian's walking mode of the described instantiation of step 4.2, the instantiation method of the implicit state of the Hidden Markov Model (HMM) of its indoor pedestrian's walking mode is: at first indoor pedestrian's walking mode is classified according to direction of travel and the speed of travel; Wherein direction of travel is usingd to 15 degree as interval, 360 degree are divided into to 24 direction sections, the angle of supposing indoor pedestrian's direction of travel is θ, and the direction section under it is so because of indoor normal walking speed scope, at 0.5m/s-2m/s, then the speed of travel being divided three classes, is respectively (0.5m/s-1m/s), middling speed (1m/s-1.5m/s), fast (1.5m/s-2m/s) at a slow speed; According to the direction of described walking and the speed of walking, walking mode can be divided into altogether 72 kinds, every kind of corresponding direction section of walking mode and a speed type.

6. the method for indoor pedestrian's walking mode identification according to claim 4 and trajectory track, it is characterized in that: the Hidden Markov Model (HMM) of the indoor pedestrian's walking mode of the described instantiation of step 4.2, the instantiation method of the observed reading of the Hidden Markov Model (HMM) of its indoor pedestrian's walking mode is: by the pedestrian is divided into to several equal time intervals at the capable time made a move, gather an observed reading every an equal time period, described observed reading is arranged in to observation vector according to time sequencing; Wherein, described observed reading is three-dimensional vectorial O=(acce, Δ hd, spd), and acce means pedestrian's z-axle acceleration, the azimuthal variation amount that Δ hd means the pedestrian, the speed of travel that spd means the pedestrian.

7. the method for indoor pedestrian's walking mode identification according to claim 4 and trajectory track, it is characterized in that: the Hidden Markov Model (HMM) of the indoor pedestrian's walking mode of the described instantiation of step 4.2, the instantiation method of the output probability of the Hidden Markov Model (HMM) of its indoor pedestrian's walking mode comprises following sub-step:

Step 4.2.1: the present invention is by capable corresponding three observational variables the method that it is averaged of making a move with each interior walking mode of multi collect indoor occupant, set up a standard fingerprint base for each walking mode, each fingerprint base comprises a walking mode s _jwith sequence of observations v _icorresponding relation, this corresponding relation is expressed as a bivector (s _j, v _i);

Step 4.2.2: a given sequence of observations v _i, search each walking mode s in fingerprint base _jcorresponding standard observation value sequence v _j;

Step 4.2.3: calculate given observed reading v _ithe standard observation value v corresponding with each walking mode _jbetween Euclidean distance;

Step 4.2.3: finally with the zero-mean Gaussian function, calculate each observed reading v _iwith corresponding state s _jbetween emission probability.

8. the method for indoor pedestrian's walking mode identification according to claim 4 and trajectory track, it is characterized in that: the Hidden Markov Model (HMM) of the indoor pedestrian's walking mode of the described instantiation of step 4.2, the instantiation method of the state transition probability of the Hidden Markov Model (HMM) of its indoor pedestrian's walking mode is: described Hidden Markov Model (HMM) state transition probability is divided into direction transition probability and speed transition probability;

The direction transition probability means that the pedestrian transfers to the probability of the residing direction of another one state from a residing direction of state, if the pedestrian transfers to state j, i from state i _heading, j _headingmean respectively the residing direction section of state i and state j, direction transition probability P _headingthe computing formula of (i, j) is as follows:

$P_{\mathrm{heading}}(i,j)=\frac{\mathrm{\θ}}{2^n},n=|i_{\mathrm{heading}}-j_{\mathrm{heading}}|,$

Wherein θ means to work as i _heading=j _headingthe time the direction transition probability of pedestrian's Hidden Markov Model (HMM);

Because of P _heading(i, j) meets constraint condition

The probability sum of the direction that likely shifts be 1, therefore the value that by this constraint condition, can calculate θ is about 0.334; If the pedestrian transfers to state j, i from state i _speed, j _speedmean respectively state i and the residing speed type of state j, speed transition probability P _speedthe computing formula of (i, j) is as follows:

$P_{\mathrm{speed}}(i,j)=\frac{\mathrm{\ω}}{2^n},n=|i_{\mathrm{speed}}-j_{\mathrm{speed}}|,$

Wherein ω means to work as i _speed=j _speedthe time the direction transition probability of pedestrian's Hidden Markov Model (HMM);

Because of P _speed(i, j) meets following constraint condition

$\underset{j=1}{\overset{k}{\mathrm{\Σ}}}{P}_{\mathrm{speed}}(i,j)=1,$

The probability sum of the speed that likely shifts be 1, therefore can calculate ω by this constraint condition, be 0.5;

, for a Hidden Markov Model (HMM) λ=(A, B, π), state transition probability matrix A is:

a _ij＝P _heading(i，j)×P _speed(i，j)，

Wherein, a _ijmean the state transition probability of Hidden Markov Model (HMM), B means the output probability matrix of Hidden Markov Model (HMM), and π means the original state probability matrix of Hidden Markov Model (HMM).

9. the method for indoor pedestrian's walking mode identification according to claim 1 and trajectory track, it is characterized in that: the walking mode that the implicit Markov model according to foundation described in step 5 and sensor observed reading identify indoor this pedestrian, its specific implementation comprises following sub-step:

Step 5.1: the sequence of observations O that inputs any walking mode _i, i=1,2 ... n;

Step 5.2: set up Hidden Markov Model (HMM) λ=(A, B, the π) of indoor pedestrian's walking mode, wherein:

π＝[π ₁，π ₂，...，π _N]；

Wherein, a _ijthe state transition probability that means Hidden Markov Model (HMM), b _j(i) mean the output probability of Hidden Markov Model (HMM), N means the status number of implicit Er Kefu model, π _jthe original state probability that means Hidden Markov Model (HMM);

Step 5.3: Hidden Markov Model (HMM) λ=(A, B, the π) for pedestrian's walking mode of setting up, calculate with viterbi algorithm the probability that all possible walking states is corresponding, then by sequence, find out maximum probability wherein:

δ ₁(j)=π _jb _j(o ₁) 1≤j≤N

δ _t(j)=max _i(δ _t-1(j)a _ijb _i(o _t)) 2≤t≤T,1≤k≤N；

Wherein, T means the time of the walking cost of this walking mode, calculates t constantly in state S _jmaximum probability δ t (j);

Step 5.4: judgement t is the state S in maximum probability constantly _jprobability δ t (j) be greater than fixing threshold value p (TH)?

If: continue to carry out following step 5.5,

If not: finish identification;

Step 5.5: state corresponding to maximum probability that record calculates at every turn, for recalling state each time; Maximum probability δ for a viterbi algorithm computation process _t-1(i) and one state transition probability a _ij, traceback information ψ _t(j) computing formula is as follows:

ψ _t(j)=arg?max _i[δ _t-1(i)a _ij] 2≤t≤T,1≤j≤N；

Step 5.6: utilize described traceback information ψ _t(j), with back-track algorithm, identify pedestrian's walking mode $q_t^{*},t=T-1,T-2,...,1,$ Its iterative process is as follows:

$q_T^{*}=S_N$

$q_t^{*}={\mathrm{\ψ}}_{t+1}\left(q_{t+1}^{*}\right),t=T-1,T-2,...,1.$

10. the method for indoor pedestrian's walking mode identification according to claim 6 and trajectory track, it is characterized in that: described p (TH) gets 0.6.

11. the method for indoor pedestrian's walking mode identification according to claim 1 and trajectory track, it is characterized in that: the employing described in step 6 abnormity point identification, deletion and bad block reparation method to the coupling after run trace revised, its specific implementation comprises following sub-step:

Step 6.1: the sequence of observations O that inputs any one walking mode _i, i=1,2 ... n; Hidden Markov Model (HMM) λ=(A, B, the π) of indoor pedestrian's walking mode of setting up; Wherein, A means the state transition probability matrix of Hidden Markov Model (HMM), and B means the output probability matrix of Hidden Markov Model (HMM), and π means the original state probability matrix of Hidden Markov Model (HMM).

Step 6.2: set a moving window that length is k for the sequence of observations, the sequence of observations that is n by length is divided into the short sequence that length is k;

Step 6.3: moving window moves a unit at every turn backward, wherein, and the number of times C=(n-k+1) that moving window moves altogether;

Step 6.4: judge whether moving window arrives sequence of observations end,

If so, monitoring finishes;

If not, continue to carry out following step;

Step 6.5: the maximum probability δ t (j) of walking mode that calculates the short sequence at this moving window place with viterbi algorithm:

δ ₁(j)=π _jb _j(O ₁) 1≤j≤N

${\mathrm{\δ}}_t\left(j\right)={\max }_i\left({\mathrm{\δ}}_{t-1}\left(j\right)a_{\mathrm{ij}}{b}_{{\mathrm{iO}}_t}\right),2\≤t\≤T,1\≤k\≤N;$

Wherein, a _ijthe state transition probability that means Hidden Markov Model (HMM), b _j(i) mean the output probability of Hidden Markov Model (HMM), N means the status number of implicit Er Kefu model, π _jthe original state probability that means Hidden Markov Model (HMM), T means the time of the walking cost of this walking mode;

Step 6.6: judgement maximum probability δ t (j) and threshold values p (TH) size of setting;

If maximum rating probability δ t (j) is greater than this threshold value, this time short sequence is normal point, returns and continues to carry out described step 6.3;

If maximum rating probability δ t (j) is less than this threshold value, this time short sequence is abnormity point, continues to carry out following step;

Step 6.7: this short sequence is carried out to abnormity point and delete and the bad block reparation, the method for its bad block reparation is first to delete original abnormity point, then passes through the method for curve point of position matching in the abnormity point of deleting, then returns and carry out described step 6.3.

Images